ESCUELA TÉCNICA SUPERIOR DE INGENIERIA (ICAI)
Instituto de Investigación Tecnológica (IIT)
Voltage Control Design of Wind Energy
Harvesting Networks
Tesis para la obtención del grado de Doctor
Supervisors: Prof. Dr. D. Enrique Lobato Miguélez Prof. Dr. D. Ignacio Egido Cortés Author: Ing. Dña. Elena Sáiz Marín
DE LA DEFENSA DE TESIS DOCTORAL
TÍTULO: Wind Energy harvesting networks voltage control design AUTOR: Elena Sáiz Marín
DIRECTOR: Enrique Lobato Miguélez, Ignacio Egido Cortés TUTOR-PONENTE: Enrique Lobato Miguélez
DEPARTAMENTO: Instituto de Investigación Tecnológica
FACULTAD O ESCUELA: Escuela Técnica Superior de Ingeniería
Miembros del Tribunal Calificador:
PRESIDENTE: Firma:
VOCAL: Firma:
VOCAL: Firma:
VOCAL: Firma:
SECRETARIO: Firma:
Fecha de lectura:
YOU VERY MUCH!
En primer lugar quería agradecerle a Luis Rouco la oportunidad que me brindó entrando al IIT, en pocos sitios se aprende tanto como aquí, y por su puesto a mis directores, Enrique Lobato e Ignacio Egido que confiaron en mi tesis hasta cuando yo no lo hacía. También a toda la gente con la que he trabajado a lo largo de estos años y de los que he aprendido un montón. Muy en especial a todos los miembros del proyecto TWENTIES el cuál representó el pistoletazo de salida de esta tesis. Juan Carlos Pérez Campión, Iñigo Azpiri, Clara Combarros, Roberto Veguillas, Juan Rivier, Placido Ostos, Miguel Lorenzo, Guillermo Juberias, Miguel Linares…. Y por supuesto a los de dentro del IIT Javier García, Pablo Frías, Camila Formoso y Mercedes VallesGracias a todos mis compañeros o mejor dicho amigos en todo este tiempo. A los que ya estaban cuando yo llegué y que siempre han estado dispuestos a ayudarme Lukas, Ale, Maria, Paco a los que han compartidos desayunos, comidas (con curso de lekue incluido), pachangas de baloncesto… Álvaro, Alessandro, Adela, Rafa, Carlos, Adrián, William y por su puestos a Inma y Luis que compartieron principio y fin de tesis siendo un soporte esencial durante la misma, qué hubiese hecho yo sin vosotros! Y como no a toda la gente que ha formado parte esencial en a mi vida. A los del Cole, la banda y por supuesto a los de la Universidad. Tampoco me quería olvidar de mi profesor de clarinete, Juan Luis, que me descubrió la melanina y me ha ayudado más de lo que él se puede llegar a imaginar.
I will never forget my time in Ireland, which also proved to be my most fruitful research period. Thanks to Andrej Gubina who suggested I make the move there and, of course, to Andrew Keane who embraced me into his research group. Thank you very much to all the ERC students who made my stay unforgettable – Álvaro, Killian (my jazz professor!), Peter (I now never miss any Ireland rugby match!), Connor, Alison, Mostofa, Mario, Padraig, James, Jonathan ... and a special and big thanks to Paul Cuffe. Thanks for all the discussions during my stay and all the Skype conversations when I was back in Spain. I have learned a world of new vocabulary and you always provided me with useful, distinctive points of view. I am certain I would not have finished my PhD without your help. Also, huge thanks to Ronald Besser and again to Killian Mc-Kenna and Paul Cuffe for proofreading the thesis and correcting my “Spanglish”. You were my saviours and I am eternally grateful.
Y por supuestísimo a toda mi familia, no se podría haber tenido más suerte. Gracias a mi hermana la mejor reportera del mundo que siempre ha sido la alegría de la casa. A mi pronto cuñado Olly (Bienvenido a los Sáiz-Marín), gracias por corregir mi inglés aunque te sonase a chino lo que decía. A mi madre, el gran pilar de la familia, que dejo de lado su camino para dedicarse por entero a sus hijas y que no te lo agradecemos lo suficiente, tu me has enseñado el valor del esfuerzo y de la perseverancia. Como pasito a pasito se alcanza cualquier cosa que nos podamos proponer. A mi padre, yo ya de pequeña tenía claro que de mayor quería ser como tu (ni austronauta, ni veterinaria, ni médico…) así que cuando alguien me preguntaba que quería ser de mayor les decía “yo, como mi padre” ese deseo me ha llevado hasta aquí y ha sido el motor de mi vida. Mil gracias sois los mejores. Y como no, a Ángel (mi futura familia) no me hubiese imaginado una persona mejor con la que compartir mi vida, el más bueno que existe y al
ABSTRACT
Specific networks developed solely to harvest wind energy (HNet) are becoming a common scheme. Moreover, it is foreseen that in the near future, they will be the solution adopted for large integration of wind, commanding greater transmission system impact potential. However, despite the huge literature available related to wind farms and their integration, few works focus on these networks, a gap covered in depth by this thesis. These potentials are highly dependent on the HNet characteristics; hence its classification is essential. This is done by analyzing the influence of OLTC transformers based on three relevant indices: PQ chart, power losses, and voltage margins deriving three different HNets types (A, B and C). For each one, the most suitable control strategy is proposed considering simple control schemes which can nowadays be implemented without additional investments. Consequently their steady-state performance and temporal evolution analysis are required. In that sense a wide variety of techniques are used throughout the thesis: data-mining techniques (regressions, clustering, decision trees …), metaheuristic algorithms (genetic algorithms and multiple particle swarm optimization), quadratic programming or multi-period OPF.
Specifically, for type A, power loss minimization strategy based on control rules is suggested allowing the understanding of power flows performance within the grid and identifying those wind farms with negligible impact. For that purpose a novel variable, active power losses from wind farm i to the transmission network bus (Plossi), was defined. In addition, to avoid online computations, the total HNet active power has been considered as an explanatory variable resembling the power factor concept (with respect a global magnitude instead of the individual wind farm active power). In that manner, simple regression rules are used to estimate wind farm reactive power, and a classification tree for each transformer is used to estimate their taps.
For type B, a minimum HNet impact on the transmission network strategy is suggested analyzing different possible control schemes: reference control, power factor control, local voltage control (representative of the current regulation direction) and remote voltage control. For each control scheme static fit-and-forget settings are obtained thanks to the AC-multi-period OPF and therefore autonomous control schemes are obtained. Comparing these control schemes it has been observed that the best option is a voltage control where wind farms control the voltage at a remote bus. Otherwise, of the localized control schemes that do not require telemetry, power factor control scheme has a better performance contrary to the widespread idea of local voltage control adequacy as many regulation proposals suggest.
Finally for type C, a pro-active voltage control (i.e., the whole HNet resembles a conventional power plant) is suggested demanding adaptive static parameters contrarily to what was proposed for the previous HNet types. Hence, a central controller distributes set-point depending on the operational and external conditions. This central controller has been developed following the guidelines of well-known TNet hierarchical voltage control and more specifically in its secondary loop performing and optimization by means of quadratic programming every 10 seconds. Nonetheless, several modifications
have been made in accordance with HNet characteristics and the nature of distributed generation such as its variability as are explained throughout the thesis.
In addition, any control scheme design should also consider the dynamic coordination of wind farms and cascade OLTC transformers to ensure that such set points redispatche can be harmoniously achieved. In that sense this thesis proposes tuning offline relevant dynamic settings (i.e., settings that affect the control scheme temporal time evolution) to be applied to Type B and C HNets. These settings are wind farms’ controller time constants and for OLTC transformers the time delay and dead band. This last setting is not commonly used with coordination purpose although there is no barrier that impedes its use, as this thesis proposes. For that purpose two well-known metaheuristics algorithms (Genetic algorithm and MOPSO) have been used. On one hand the former provides the settings tendency whereas the latter one provides the whole Pareto frontier; allowing the settings categorization depending on the agents preferences (tap changes minimization, voltage breaches minimization and voltage deviation minimization). Concerning type B, this analysis reinforces the idea of inadequacy of local voltage control scheme owing to the necessity of slow controller action for avoiding oscillations. Finally, a demanding voltage control such as the remote one significantly increases the number of tap changes. Concerning type C, the same method has been employed focusing this time only on the MOPSO algorithm. The results obtained have been expanded clustering the Pareto front obtaining different dynamic settings patterns.
1.1.MOTIVATION AND SCOPE ... 1
1.2.WIND ENERGY HARVESTING NETWORKS ... 3
1.3.CURRENT SCENARIO OF WIND VOLTAGE CONTROL ... 5
1.3.1. Technological development ... 6
1.3.2. Regulation framework of the wind energy voltage control provision ... 8
1.4.OBJECTIVES ... 10
1.4.1. Steady-state assesment and HNets classification ... 11
1.4.2. Optimal static settings design ... 12
1.4.3. Time variant and offline tuning of dynamic settings ... 12
1.5.STRUCTURE OF THE DOCUMENT ... 12
2. STATE OF THE ART ... 15
2.1.CONTROL CONCEPTS ... 16
2.1.1. Changing wind farms passive paradigm ... 16
2.1.2. Communication Schemes ... 17
2.1.3. Control configuration ... 17
2.1.4. Coordination ... 19
2.1.5. Control parameters ... 22
2.1.6. Global overview ... 22
2.2.DNET AND SMART GRIDS SOLUTIONS ... 23
2.2.1. OLTC Transformers coordination ... 23
2.2.2. Enhanced utilization of reactive power capabilities, Steady-state coordination ... 24
2.2.3. Dynamic coordination ... 26
2.3.TRANSMISSION NETWORK HIERACHICAL CONTROL SCHEMES ... 31
2.3.1. Italian hierarchical control ... 31
2.3.2. French hierarchical control ... 33
2.3.3. Hierarchical control not implemented ... 35
2.3.4. Incorporation of wind farms in transmission networks, PQ chart ... 38
2.4.FIELD EXPERIENCES AND INTERNATIONAL PROJECTS ... 39
2.5.SUMMARY AND GAPS TO BE ADDRESSED ... 42
3. WIND ENERGY HARVESTING NETWORKS CLASSIFICATION BASED ON THEIR POTENTIAL ... 49
3.1.VOLTAGE CONTROL PROVISION BENEFITS ... 49
3.2.DETAILED CHARACTERIZATION OF HNETS POTENTIAL ... 51
3.2.1. PQ chart and control schemes... 52
3.2.2. Test networks ... 54
3.2.4. Minimizing power losses ...57
3.2.5. HNet limits and potential and comparion with current and proposed schemes ...58
3.3.STRATEGY SELECTION ...63
3.4.SUMMARY AND CONCLUSIONS ...65
4. MINIMIZATION OF POWER LOSSES BY ONLINE CONTROL RULES ...67
4.1.METHODOLOGY ...68
4.1.1. Optimal operational scenarios building ...68
4.1.2. Explanatory variables ...69
4.1.3. Development of control rules ...72
4.1.4. Output obtained and implementation ...73
4.2.EXAMPLE CASE ...74
4.2.1. Data base building ...74
4.2.2. Regression rules in the event of not considering OLTC transformers ...74
4.2.3. Optimal voltage control with transformers’ taps ...77
4.3.DETAIL POWER LOSSES IMPACT ...81
4.4.SUMMARY AND CONCLUSIONS ...84
5. MINIMIZATION OF HARVESTING NETWORK IMPACT ON THE TRANSMISSION NETWORK ...87
5.1.METHOD OVERVIEW ...87
5.2.STEADY STATE ANALYSIS ...91
5.2.1. AC multi-period OPF ...91
5.2.2. Results and discussion...93
5.3. DYNAMIC ANALYSIS ...98
5.3.1. Objective to be minimized ...99
5.3.2. Tuned settings ...101
5.3.3. Initial vs tuned settings...102
5.4.SUMMARY AND CONCLUSIONS ...103
6. PRO-ACTIVE VOLTAGE CONTROL ...105
6.1.METHOD OVERVIEW ...106
6.2.CENTRAL CONTROLLER COMPONENTS ...107
6.3.HNET SECONDARY CONTROL LOOP ...108
6.3.1. proposed Formulation ...108
6.3.2. Trade-off among objectives determination ...112
6.3.3. First order lag incorporation and control adequacy ...114
6.4.OLTCTRANSFORMERS ...117
6.5.1. Scheme performace ... 118
6.5.2. Dynamic coordination ... 119
6.5.3. Non dominated solutions analysis ... 120
6.5.4. Comparison ... 121
6.6.SUMMARY AND CONCLUSIONS ... 122
7. CONCLUSIONS AND FUTURE RESEARCH ... 123
7.1.CONCLUSIONS ... 123
7.2.CONTRIBUTIONS ... 124
7.2.1. Steady-state assessment ... 125
7.2.2. Classification of harvesting networks ... 125
7.2.3. Control rules algorithm to minimize power losses ... 126
7.2.4. Control scheme static settings and temporal performance ... 127
7.2.5. Offline tuning of dynamic settings ... 127
7.3. PUBLICATIONS ... 128
7.4.FUTURE RESEARCH ... 129
7.4.1. Replicability and scalability ... 129
7.4.2. HNet integration into a TNet Hierarchical voltage control ... 129
7.4.3. Experimental validation of the control schemes proposed ... 129
APPENDICES
A. TEST NETWORKS EVALUATED ...145
B. QV RELATIONSHIP AND LOCAL HOSTING CAPACITY ...149
B.1 VOLTAGE RISE EFFECT ...149
B.2 LOCAL HOSTING CAPACITY ...150
B.3 SPANISH LOCAL CAPACITY EVALUATION ...156
B.4 MATHEMATICAL DEMONSTRATION ...160
C. SENSITIVITY MATRIXES ...163
C.1HNET LINEAL MODEL ...163
C.2SENSITIVITY MATRIX USED ...165
D. SIMULATORS ...167
D.1HNET SIMULATOR ...167
D.2WIND FARM SPECIFIC SIMULATOR ...169
E. WIND FARMS SCENARIOS ...171
E.1CORRELATION WIND FARM DATA ...171
E.2WIND FARMS TIME EVOLUTION PATTERNS ...173
F. METAHEURISC ALGORITHMS ...177
F.1GENETIC ALGORITHM ...177
F.2MOPSO ...179
G. DETAIL COMPARISON AMONG CONTROL SCHEMES ...183
G.1CONTROL SCHEME PERFORMANCE ...183
G.2DYNAMIC SETTINGS TUNING ...188
G.2.1. Genetic algorithm Penalty factors selected ...188
G.2.2. Dynamic settings tendency, Genetic algorithm results ...190
G.2.3. Tuned vs initial dynamic settings ...192
G.3.MARGIN VISUALIZATION ...195
H. NECESSITY OF ADAPTIVE SETTINGS AND ADEQUACY OF THE CONTROL SCHEMES PROPOSED ...197
H.1.NECCESITY OF ADAPTING THE SETTINGS IN REAL TIME OPERATION ...197
H.2.TNET VOLTAGE CONTROL ADEQUACY ...200
I. OLTC TRANSFORMERS STRATEGY FOR MAXIMIZING THE REACTIVE POWER DELIVERED BY WIND FARMS ...203
Symbol Description ACD Adaptive critic design
ANN Artificial neural network AVR Automatic voltage regulator
CART Classification and regression tree algorithm
CECRE Renewable energy control center (operated by REE) CORE Renewable energy control center (operated by Iberdrola) CSVC Coordinated secondary voltage control
DFIG Doubled fed induction generator DG Distributed generation
DSO Distribution system operator
DNSGA-II Dynamic multiple objective genetic algorithm
ENTSO-E European network of transmission system operators for electricity FACTS Flexible alternative current transmission system
GA Genetic algorithm HNet Harvesting network
MOPSO Multiple objective particle swarm optimization MPC Model predictive control
MVMO Mean-Variance Mapping Optimization NSGA-II multiple objective genetic algorithm
OLTC On load tap changing OPF Optimal power flow
Pcc Point of common coupling Pf Power factor
P.O. Operation procedure (Spanish nomenclature) PSO Particle swarm optimization
RD Royal decree
REE Red eléctrica de España (Spanish transmission system operator)
REPORT Microprocessor reactive and voltage regulator employed in the Italian secondary voltage control
SART Sistema Automatico per la Regolazione di Tensione (Automatic system for voltage regulation)
SCADA Supervisory Control and Data Acquisition
SONI Independent electricity Transmission System Operator and Market Operator in Ireland
SVR Step voltage regulator TNet Transmission network
TSO Transmission system operator
TWENTIES
Transmission system operation with large penetration of Wind and other renewable Electricity sources in Networks by means of innovative Tools and Integrated Energy Solutions
Common symbols used throughout the thesis P Active power (MW)
Pcc Common coupling bus, connects the HNet to the TNet Q Reactive power (MVAR)
POWER LOSSES
Symbol Description
c0, c1, c2 Polynomial coefficients
fWF Active power losses factor assigned to a certain wind farm
PgrossWF Gross active power losses delivered by a certain wind farm (MWh) PnetWF Net active power losses delivered by a certain wind farm (MWh)
PWF Individual wind farm active power (MW, p.u.)
𝑄𝑊𝐹𝑚𝑖𝑛,𝑄𝑊𝐹𝑚𝑎𝑥 Maximum a minimum reactive power provided by wind farms (MVAR, p.u.) R2 Coefficient of determination
𝑆𝑊𝑚𝑖𝑛, 𝑆𝑊𝑚𝑎𝑥 Maximum a minimum wind speed (m/s)
VWF Voltage at wind farm grid connection point (kV, p.u.)
WF Wind farm
Explanatory variables selected
Ploss ij Active power losses from bus i to bus j (MW, p.u.)
Ploss WF Active power losses from wind farm WF to the transmission network bus (MW, p.u.) TQlosses Total wind energy HNet reactive power losses (MVAR, p.u.)
TP Total wind energy HNet active power (MVAR, p.u.) TPlosses Total wind energy HNet active power losses (MVAR, p.u.)
VTNet Transmission network voltage (kV, p.u.) Control variables
Qref , QWF Wind farms reactive power references (MVAR, p.u.) t Tap position
VOLTAGE CONTROLS SCHEMES
Symbol Description
K Proportional control scheme slope R Proportional control scheme droop (%) Vmeasured Measured voltage at a certain bus (kV/p.u.)
Vpcc,t Measured voltage at Pcc for period t (kV/p.u.) Vsetpoint Voltage set-point at a certain bus (kV/p.u.)
Hierarchical voltage control schemes Sets
c Controlled bus (corresponding to those buses in which a generator is located) p Pilot buses
s Critical buses
Parameters
a, b, c Coefficients of straight lines used for modelling the reactive power limits k Sample time of the discrete controller
𝑄𝑐𝑚𝑖𝑛, 𝑄𝑐𝑚𝑎𝑥
Minimum and maximum value of the reactive power that can be provided by the generators (MVAR, p.u.)
Tc Time constant of the secondary voltage control (s) wv Weight of the pilot buses voltage deviation term wq Weight of the reactive power increment at the generator wg Weight of the voltage increment at the generator
Symbol Description
𝑉𝑐𝑚𝑖𝑛, 𝑉𝑐𝑚𝑎𝑥 Minimum and maximum value of the voltage of the controlled buses 𝑉𝑝𝑚𝑎𝑥 Maximum pilot bus reference
𝑉𝑠𝑚𝑖𝑛, 𝑉𝑠𝑚𝑎𝑥 Minimum and maximum value of the voltage of the critical buses
𝛼
Coefficient between zero and one that considers the dynamic performance. The time constant of the control fulfill the following formulas as demonstrated [Pagola 1993]. Hence, it can be elucidated that if 𝛼 = 1 a dead beat control is obtained. This means that the control action is achieved in just a sample time ∆𝑡 (considering perfect
modelling of the system and a lineal response) 𝑇𝑐= −∆𝑡 ln (1− 𝛼)
γ Gain
Input of the controllers
𝐴𝑐𝑐(𝑘)
Sensitivity matrix of the reactive power generated with respect to the voltage at the
controlled buses both correspond to the generator buses 𝜕𝑄𝑐
𝜕𝑉𝑐 in sample k (MVAR/kV)
𝑄𝑐(𝑘) Reactive power vector of the controlled buses in sample k (MVAR, p.u.)
𝑆𝑝𝑐(𝑘)
Sensitivity matrix of the pilot buses voltages with respect to the controlled voltages 𝜕𝑉𝑝
𝜕𝑉𝑐 in sample k. In the specific case that is addressed in the thesis a single pilot bus is considered. Hence, a vector instead of a matrix is evaluated. To outline this
distinction a different notation is used v𝑆𝑝𝑐(𝑘) 𝑆𝑠𝑐(𝑘)
Sensitivity matrix of the critical buses voltages with respect to the controlled voltages 𝜕𝑉𝑝
𝜕𝑉𝑠 in sample k
𝑉𝑐(𝑘) Voltage vector of the control buses in sample k (kV, p.u.) 𝑉𝑝(𝑘) Voltage vector of the pilot buses (kV, p.u.)
𝑉𝑝
𝑟𝑒𝑓 Voltage vector of the references of the pilot buses provided by the higher control
loop (kV, p.u.)
𝑉𝑠(𝑘) Voltage vector of the critical buses in sample k (kV, p.u.) ∆𝑉𝑐𝑟𝑒𝑓(𝑘) Voltage reference increment (kV, p.u.)
Output of the controllers
𝑄𝑐 𝑟𝑒𝑓
(𝑘 + 1) Reactive power references vectors of the control buses in sample k +1 (MVAR, p.u.)
𝑉𝑐 𝑟𝑒𝑓
(𝑘 + 1) Voltage references vectors of the control buses in sample k +1 (kV, p.u.)
𝑑(𝑘) Vector of area reactive power level (%)
∆𝑇𝑙𝑜𝑠𝑠𝑒𝑠(𝑘) Active power losses component (MW) ∆𝑇𝑄𝑚𝑎𝑟𝑔𝑖𝑛𝑠(𝑘) Reactive power margins components
Figure 1-1 Current Voltage hierarchical voltage control on the TNet and the HNet
integration ... 3
Figure 1-2 Wind Energy harvesting network diagram ... 5
Figure 1-3 Reactive power capabilities ... 7
Figure 1-4 Droop controls ... 10
Figure 2-1 Control strategies and configuration ... 23
Figure 2-2 Italian hierarchical voltage control scheme. Source:[Corsi et al. 2004a] ... 32
Figure 2-3 Reactive power limits consideration. Source [de la Fuente 1997] ... 37
Figure 2-4 HNets tested in the demonstration. Source [Azpiri et al. 2013] ... 40
Figure 2-5 Communication infrastructure. Source [Azpiri et al. 2013] ... 41
Figure 2-6 Performance of the demonstration. Source [Azpiri et al. 2013] ... 42
Figure 2-7 Strategies evaluated throughout the thesis ... 44
Figure 3-1 Power system with two areas. Source [Kundur 1994] ... 50
Figure 3-2 Flow-chart implementation of the different controls ... 54
Figure 3-3 OLTC transformers impact on the PQ chart in HNet1 (a) and in HNet2 (b) 56 Figure 3-4 Voltage limitations ... 56
Figure 3-5 Impact of considering different wind farms active power productions. PQ chart in the event of taking into account transformers’ taps as control variable ... 57
Figure 3-6 OLTC transformers impact on the optimal control (minimizing power losses) in HNet1 (a) and in HNet2 (b) ... 58
Figure 3-7 Optimal voltage control comparison for HNet1 (a) and for HNet2 (b) ... 59
Figure 3-8 Absolute value of power losses for HNet1 and HNet2... 60
Figure 3-9 Current situation versus optimal voltage control (minimizing power losses) HNet1(a) and HNet2 (b) ... 60
Figure 3-10 Maximum proportional control for HNet1and HNet2 ... 62
Figure 3-11 Limit performance of several HNet ... 64
Figure 3-12 HNet characterization ... 65
Figure 4-1 Data base building process ... 69
Figure 4-2 Diagram of a radial HNet that connects three different wind farms to the transmission network bus (TNet) ... 71
Figure 4-3 Adequacy of the explanatory variable ... 72
Figure 4-4 Control proposed using quadratic regression functions for each wind farm, in the event that only the reactive power capabilities of wind farms are used as control variables. ... 73
Figure 4-5 Lagrange multipliers of the equality constraint on the reactive power demand in each wind farm. ... 75
Figure 4-6 Relation between the reactive power and the individual active power of each wind farm... 76
Figure 4-7 Relation between the reactive power and its voltages ... 76
Figure 4-8 Relation between the reactive power and the total active power... 77
Figure 4-9 Wind farm reactive power vs voltage measured at its terminals ... 78
Figure 4-10 Wind farm reactive power vs total active power. ... 78
Figure 4-11 Example of a decision tree obtained for estimating the tap position of a specific OLTC transformer... 79
Figure 4-12 Confusion matrix. (a) 132kV/20kV transformer (b) 400kV/132kV
transformer ... 80
Figure 4-13 Confusion matrix 400kV/132kV transformer incorporating the transmission network voltage as explanatory variable... 80
Figure 4-14 Validation of the online control scheme proposed ... 81
Figure 4-15 Gross and net power, HNet and wind farms power losses ... 82
Figure 4-16 power losses in reference control scheme — power losses in power losses minimization control scheme in megawatt-hour... 83
Figure 4-17 Power losses minimization scheme vs proportional control scheme ... 83
Figure 4-18 Reactive power provided by each wind turbine located in WF1. ... 84
Figure 5-1 A schematic view of the complete proposed approach applied. Type B ... 88
Figure 5-2 Reference and power factor control ... 89
Figure 5-3 Local and remote voltage control ... 90
Figure 5-4 Active power scenarios ... 92
Figure 5-5 Multi-period scenarios evaluated ... 93
Figure 5-6 PV characteristic of the HNet ... 95
Figure 5-7 PQ characteristic of the HNet at the Pcc ... 96
Figure 5-8 QV characteristic of the HNet at the Pcc ... 96
Figure 5-9 PQ characteristic of the local voltage control ... 97
Figure 5-10 PQ characteristic of the remote voltage control ... 98
Figure 5-11 Remote voltage control dynamic performance ... 99
Figure 5-12 Comparison of non-dominated solutions: Reference, power factor and remote voltage control schemes ... 102
Figure 5-13 The remote voltage control scheme. Initial settings vs final settings ... 103
Figure 6-1 A schematic view of the complete proposed approach applied. Type C ... 106
Figure 6-2 Projection onto the columns space of a 3 by 2 matrix [Strang 1988] ... 109
Figure 6-3 Reactive power assignment ... 111
Figure 6-4 Determination of reactive power maximization weight ... 113
Figure 6-5 Time domain simulations considering different W values ... 114
Figure 6-6 Sample time impact ... 115
Figure 6-7 Discrete control system ... 115
Figure 6-8 Simplified system analyzed... 116
Figure 6-9 Central controller performance for a random set of non optimal dynamic settings ... 119
Figure 6-10 Search space (indicating the solution corresponding to the initial set of dynamic settings) and non-dominated solutions ... 120
Figure 6-11 Non-dominated solutions characterized in different clusters ... 120
Figure 6-12 Temporal evolution simulation. Improvements due to dynamic settings tuning ... 122
Figure A-1 First wind energy HNet evaluated (HNet1) ... 146
Figure A-2 Second wind energy HNet evaluated (HNet2). ... 146
Figure A-3 PQ chart at wind farms connection bus [REE 2011] ... 147
Figure A-4 QVchart at wind farms connection bus [REE 2011, ENTSO-E 2013] ... 148
Figure B-1 Simplified network model of both HNet and TNet ... 150
Figure B-3 Voltage for two different X/R ratios (SCC=4000 MVA, VTNet=1.00 p.u.) in
two situations ... 152
Figure B-4 Wind turbines extrapolated PQ chart ... 152
Figure B-5 Reactive power for two different ratios (SCC=4000 MVA, VTNet=1.00 p.u.) in two situations ... 153
Figure B-6 Reactive power and voltage for a ratio equal to 2 and different SCC when the wind farm provides voltage control ... 154
Figure B-7 Voltage for two different VTNet (1.00 p.u. and 1.04 p.u.) maintaining the SCC and the X/R ratio at a fixed value (SCC=4000 MVA X/R=6) ... 154
Figure B-8 Summary of the hosting capacity limit without control and the increment of the hosting capacity thanks to the voltage control provision ... 155
Figure B-9 Statistical data of SCC and X/R ratio 400 kV buses within the Spanish system (year 2010)... 156
Figure B-10 Statistical data of SCC and X/R ratio 220kV buses within the Spanish system (year 2010)... 157
Figure B-11 Comparison of simplified and detailed models of TNet ... 160
Figure D-1 Simulator used for evaluating the performance of each control ... 168
Figure D-2 Internal wind farm simulator ... 169
Figure E-1 Wind farms distribution function. ... 172
Figure E-2 Process of creating scenarios ... 172
Figure E-3 Active power profile categorization ... 174
Figure E-4 Active power profile selected ... 174
Figure E-5 Shape clusters ... 175
Figure F-1 Behaviour of one particle base on its known best experience ... 179
Figure F-2 Evolution of wind farms control time constant of the reactive power control of wind farms (Local and remote voltage control) ... 180
Figure G-1 Reference control scheme dynamic performance ... 184
Figure G-2 Power factor control scheme dynamic performance ... 185
Figure G-3 Local voltage control scheme dynamic performance ... 186
Figure G-4 Impact of the proportional slope on the dynamic performance ... 187
Figure G-5 Impact of wind farms’ time constants... 187
Figure G-6 Time delay and dead band for the reference control in the event that just one term of the objective function is minimized ... 189
Figure G-7 Fitness evolution for different controls evaluated ... 191
Figure G-8 Transformers time delay for the different controls evaluated ... 191
Figure G-9 Transformers dead band (Transformer and Power factor control) ... 192
Figure G-10 Evolution of wind farms control time constant of the reactive power control of wind farms (Local and remote voltage control) ... 192
Figure G-11 Reference control scheme. Initial settings vs final settings ... 193
Figure G-12 Power factor control scheme. Initial settings vs final settings ... 194
Figure G-13 Local voltage control scheme. Initial settings vs final settings ... 194
Figure G-14 Tuned schemes comparison ... 195
Figure G-15 Algorithm comparison. Reference control... 196
Figure G-16 Algorithm comparison. Power factor control ... 196
Figure H-1 Search space of reference control (Chapter 5), remote voltage control
(Chapter 6) and central controller ... 199
Figure I-1 wind farm grid voltage impact (simplified model) ... 204
Figure I-2 Reactive power depending on the wind turbine feeder location ... 204
List of tables
Table 1-1. Reactive power technical requirements ... 9 Table 2-1. References summary ... 46 Table 3-1. Hosting capacity increase by means of voltage control provision ... 51 Table 3-2. HNet characteristics summarize ... 52 Table 4-1. R2of wind farms without transformers’ taps ... 77 Table 4-2. R2of wind farms in the event of considering OLTC transformers ... 78 Table 4-3. Decision tree performance. Percentage bad classified ... 80 Table 5-1. Variables optimized in the different controls schemes... 91 Table 5-2. Steady state settings considering HNet2 ... 94 Table 5-3. Sensitivity analysis of the local voltage control ... 98 Table 5-4. Final dynamic settings optimized for the different control schemes ... 101 Table 6-1. Settings to be optimized ... 106 Table 6-2. Characteristics of the central controller proposed ... 108 Table 6-3. Dynamic settings patterns... 121 Table A-1. Installed active power of each wind farm within HNet1 ... 145 Table A-2. Installed active power of each wind farm within HNet2 ... 145 Table B-1. Spanish TNet statistical data ... 157 Table B-2. Summary of wind power limitations (MW) considering mean ... 158 Table B-3. Summary of wind power limitations (MW) considering minimum values 158 Table B-4. Hosting capacity increment comparison ... 159 Table G-1. Objective to be minimized and dynamic settings to be tuned ... 188 Table G-2. Terms evaluated depending on the optimized term ... 190 Table H-1. Optimal settings for different values of Vsetpoint for reference and power factor control ... 198 Table H-2. Optimal settings for different values of Vsetpoint for local and remote voltage control ... 199 Table H-3. Secondary voltage control formulation ... 201
CHAPTER 1
1.
INTRODUCTION
Before explaining the thesis itself, this chapter presents the problem stament and the derived specific objectives. Specifically, it is structured as follows. The motivation and scope of the thesis is outlined in section 1.1. As the title states, the thesis focuses on wind energy harvesting networks possessing the characteristics and peculiaritites presented in section 1.2. Subsequently, the current situation of wind generation, from technical and regulatory perspectives is presented in section 1.3. Then, the objectives, which are the main questions that are answered in this thesis, are defined in section 1.4. Finally, the structure of the whole document is given in section 1.5.
1.1.
MOTIVATION AND SCOPE
Recent years have seen a global significant growth in the use of wind power. For instance, Spain (which is used as a case example throughout the thesis) has the fourth highest installed capacity of wind power, with 22,959 MW at the end of year 2013 [Global Wind Energy Council 2013]. The level of penetration1 can now exceed 60% on certain windy days; for example, on the 24th of September 2012 it was 64.2% at 03:00. In addition, the Spanish Ministry of Industry, Tourism and Trade regards as a probable scenario 34,318 MW of installed on-shore wind power capacity and 750 MW of
shore wind power capacity by 2020 [Spanish government, 2011]. This wind penetration level could be reached in addition to other factors, because the wind generation is already contributing to the voltage control in a restricted way. As will be seen later, the Spanish operator currently requires a tight power factor range. Within that range, most wind farms are operated at a lagging power factor. The associate reactive consumption is essential during off-peak hours when the voltage raise is very significant due to the reactive generation of unloaded lines. Moreover, it should not be forgotten that is precisely during those hours when typically there is more wind available. In order to avoid this situation several reactors are connected. However, it has been seen that these elements sometimes are not enough and between 70 and 80 lines have been already disconnected from the grid in order to solve this situation.
Given these trends, increasing the controllability of these energy resources is becoming a necessity. In fact, the Spanish transmission system operator has already proposed a new grid code which demands voltage control functionality from wind farms, such that the reactive power delivered by the wind farm depends on the voltage deviation at its connection point [REE 2011]. It is clear that better harnessing of the reactive power from distributed generators has much to offer for the broader power system, both from the steady-state and dynamic perspectives. However, the various benefits that could be obtained with different control schemes should be analyzed in detail in order to understand the relative strength of each approach.
This thesis deals with this issue focusing on wind energy harvesting networks (HNets). These networks, which are explained in section 1.2, were developed solely to harvest energy and thus, no demand customers are accommodated. In such networks the internal voltage profile is principally established by the bulk supply tap-changing transformers that connect the HNet to the transmission system. Therefore, some questions arise: could the introduction of wind farm voltage control cause unwanted interactions with current transformer voltage control? How might both controller types be coordinated to avoid unnecessary tap changes and possible instabilities?
It should be noted that voltage control provision in the transmission network (TNet) by decentralized elements such as generators and OLTC transformers is not a new concept and, although the wind generation peculiarities should be undertaken, the philosophy behind the HNet control is equivalent to the existing one. In the TNet several hierarchical voltage controls are currently being operated or have been evaluated (e.g., France [Lefebvre, et al. 2000], Italy [Corsi et al. 2004a, Corsi et al. 2004b], Spain [de la Fuente 1997, Alonso 2001], South Africa [Corsi, et al. 2010]). In all these hierarchical schemes different control loops can be identified: primary, plant (not always), secondary and finally tertiary, having the settling times 1s-2s, 5s-10s, 1minute and 15 minutes respectively. Although these loops are explained in more detail in Chapter 2, a brief summary of the motivation for the implementation of these schemes follows. These loops allow the complex voltage control issue to be simplified by spatial and temporal decomposition. The fastest action corresponds to the primary loop in which the AVR of the synchronous generators compensate for local voltage deviations. Then, the plant control is in charge of coordinating the action of all generators allocated in the same plant
for maintaining a desired voltage set-point and reactive power balance between generators. Those voltage set-points are calculated by the secondary control in order to compensate the voltage deviation of selected pilot buses (controlled buses). Finally, the tertiary loop determines the voltage set-point of the pilot buses as well as the action of the discrete elements such as OLTC transformers, shunts and reactors. In this last loop is where traditionally the optimal operation (minimizing power losses, maximizing reactive margins) is carried out. If wind farms are to be integrated in the existing hierarchical voltage control of a TNet, the whole HNet should behave as closely as possible to a conventional plant.
However, as depicted in Figure 1-1, the equivalent plant control of a HNet in order to avoid interactions involves more loops than the traditional one. Indeed, the HNet itself can be seen as a TNet in miniature. Firstly, wind farms internal control which is equivalent to the TNet plant control of synchronous machines. Nevertheless, two more loops should be taken into account: the HNet wind farms coordination and finally the OLTC transformers and optimization control. Thus, a question arises at this point: Do the benefits obtained thanks to incorporating wind power HNets into the TNet hierarchical voltage control justify the increase of control complexity?
Figure 1-1 Current Voltage hierarchical voltage control on the TNet and the HNet integration
The answer to this question depends on the characteristics of the HNet which affect the impact that the HNet could have in the TNet. Thus, the features of the HNet must be known before any control is proposed. In this dissertation, these properties are evaluated in accordance with the maximum reactive power that can be consumed from or generated into the TNet, the maximum voltage increase or decrease that the reactive limits imposed and finally the power losses originated within the HNet. Thanks to that preliminary analysis it is possible to identify which HNets can play an important role in the TNet control. For others HNets different strategies are investigated and proposed in the context of this thesis.
1.2.
WIND ENERGY HARVESTING NETWORKS
When the first wind farms were installed in various power systems, the usual scheme was that their power was harvested through existing distribution networks. However, those networks were not designed with that purpose. With the increment of distributed generation the power flows sometimes became bidirectional, increasing the complexity
Tertiary
Secondary
Plant
Primary
OLTC transformers
HNet wind farms coordination
Wind farms internal
control
Wind turbines
T
N
e
t
H
N
e
of network operation. This fact motivated several research efforts, for example in [Borghetti 2012] the best radial configuration in order to minimize power loss is evaluated, or, in [Gao et al. 2011] how line drop compensation of the transformers should be configured depending on the direction of the power flow.
Nowadays, due to the significant increase of wind power, another scheme is that wind farms harvest their power through a dedicated network in which there are no demands. For instance, in Spain this scheme was incentivized by the previous regulations already abrogated RD 436/2004 [Spanish government 2004] and RD 661/2007 [Spanish government, 2007]. In these regulations it was indicated that in order to get the premium the maximum power of wind farms should be 50 MW. This fact fostered the creation of a network in which several small wind farms (less than 50 MW) were located instead of developing a large one. Moreover, a common scenario is that there are not distribution networks in the areas with good wind resources and hence, a dedicated network needs to be built. As a result, in Spain approximately 85% of wind generation is allocated in harvesting networks, representing a total capacity of 19,376 MW. In addition it is also becoming an important option in Ireland [Smith, et al. 2010, Cuffe et al. 2012a]. Nevertheless, this option is far from being the most common in other countries. Owing to this fact, there is little research literature that tackles wind energy harvesting networks.
Thanks to the existence of a dedicated harvesting network the performance of the whole network resembles a conventional plant, being this fact the principal reason of why this thesis focuses on harvesting networks. In Ireland, as a case example, the grid code imposes different requirements to wind farms located in the distribution networks that to the ones directly connected to the transmission network. In the latter ones, wind farms are required to provide a similar voltage control as the conventional plants do. However, it must be noted that the philosophy behind the voltage control design is independent of the existence of demands, a fact that is discussed throughout the thesis.
Next, in Figure 1-2 a simplified diagram of an invented HNet which compromises four different wind farms is presented depicting also the internal wind farms grid. It must be noted that wind farms grid typically is radial and could be formed of a single or several feeders being all wind turbines connected to the grid through a transformer (30-20kV/0.69 kV). Those transformers have not been depicted because they are not allowed to change their taps on load. In contrast, the harvesting network could be meshed. In addition, within the harvesting network several OLTC transformers are located controlling the low voltage side as is represented in the figure with a red dot. Within these transformers two different types can be identified. On one hand, the transmission transformer, (400kV/220kV-132kV). On the other hand, the transformers that connect wind farms to the HNet (220kV-132kV/30-20kV). Hence, it can be noted that in those networks there are already control elements which should be coordinated with the control provided by wind farms as had been said. In this simplified diagram the transmission network has been represented with the Thevenin equivalent in accordance with its short-circuit power and transmission network voltage. This simplification is used throughout the thesis.
Another important aspect is that usually different wind farm owners create a joint venture for building a harvesting network. Hence, the wind farm grid could belong to different wind farm owners, e.g., companies A, B and C. Nevertheless, who is responsible for managing the harvesting network itself? The answer is none of the aforementioned wind farm owners owing to the transmission system operator forces the existence of a single distribution operator. Hence, TSO, DSO and wind farm owners should cooperate in order to maximize the advantages (for all parties involved) of reactive power capabilities.
As will be seen in subsection 1.3.2 in Spain, wind farms’ voltage set-points are sent to the meter point which varies in location: At the low voltage side if different wind farm owners share the transformers or at the high voltage side if all wind farms belong to the same owner. Hence, as can be seen in Figure 1-2 in accordance with the current grid code proposal several devices will control the same bus.
Figure 1-2 Wind Energy harvesting network diagram
Finally, it should be highlighted that this thesis focuses on onshore HNet. Offshore HNets will be commonly connected to the transmission network through a high voltage direct current (HVDC) link [Bresesti et al. 2007] and hence, appropiate analysis should be carried out. Nonetheless the philosophy of all control schemes proposed and their methodologies are totally applicable to these grids.
1.3.
CURRENT SCENARIO OF WIND VOLTAGE CONTROL
Within this subsection the current technological development and regulatory framework are discussed in subsection 1.3.1 and 1.3.2 respectively.
Transmission network
Wind energy harvesting network
Wind farm grid
RThev+j XThev
Pcc Point of Common Coupling
Meter point
Controlled bus
TSO
DSO
Wind farm owner A
A
C B
1.3.1.TECHNOLOGICAL DEVELOPMENT
The HNets compromises two different control devices: wind turbines and transformers. The characteristics of these devices are as follows:
Wind turbines: At the beginning of the development of wind energy technology, the wind turbines were of fixed speed. That means that the rotor wind turbine speed is determined by the supply grid frequency, the gear ratio and also the design of the machine. Consequently the rotor speed is independent on the wind speed [Ackermann 2005]. Normally, those wind turbines were equipped with an induction generator that was directly connected to the grid, and typically, a capacitor bank was added in order to reduce the reactive power consumption. Although this turbine has clear advantages such as its simplicity and robustness, it has important drawbacks. On one hand, it requires additional elements (capacitor banks) for providing reactive power. On the other hand, those machines are designed to achieve maximum efficiency at a certain wind speed.
Thus, in order to enhance its controllability the variable speed turbines were introduced. In fact, in Spain at the end of year 2009 the variable speed turbines represented 74.07% of the total installed turbines and, more importantly, the new ones are of variable speed. As can be derived from its name, variable speed, the main characteristic of these turbines is that they are designed to achieve maximum efficiency over a wide speed range. This fact is achieved adapting the rotor speed to the wind speed, thanks to the incorporation of a power converter. Within the variable speed turbines there are different types [Ackermann 2005]. The most relevants are the full converter and the doubly fed induction machine (DFIG). Currently, DFIG is the most common one. In this type of turbine the stator is directly connected to the grid whereas the rotor is connected to the grid through a converter allowing the provision of frequency2 and voltage control. [Ackermann 2005, Engelhardt et al. 2011, Martinez et al. 2011, Singh, et al. 2010, Xiangyu Zhang, et al. 2010, Konopinski et al. 2009, Ozturk, et al. 2009]. From the voltage control perspective (control under study in this thesis), this turbine is able to provide reactive power without adding any element. The side of the converter connected to the stator compensates the reactive power generated by the machine and consumes or generates the reactive power in accordance with the grid codes requirements. Hence, the reactive generation/consumption is limited by the power converter. Normally, owing to economic reasons, it is
2 The rotor speed is principally determined by the supply grid frequency. In the event of an induction
machine the rotor frequency (f2) is related to the supply grid frequency (f1) through the following equation f2 = (1- wr) · f1being wr the machine speed. Since the system frequency is fixed (50 Hz), the machine speed depends on f.
sized at around 25–35% of the generator rating. This fact also means that the speed range is between ± 25-35% [AEMO 2013].
In a full converter machine, the power converter has the same rating (or greater) as the generator, which is not directly connected to the grid. As a result, the generator can be operated at any speed from zero to maximum, and provides an improved reactive power capability range (PQ chart) compared to the DFIG as can be seen in Figure 1-3. In this figure a generic full converter (considering the limitation factors: converter current limitation, converter voltage limitation and active power limitation [Valverde, et al. 2014]) is depicted jointly with the common GAMESA wind turbine G87 [Gamesa 2013]. The characteristics of GAMESA wind turbine have been employed throughout the whole thesis. In addition, the extended PQV charts of both machines are compared with the synchronous machine in [Valverde, et al. 2014]
Figure 1-3 Reactive power capabilities
Finally, it must be outlined that new developments related to improving wind turbines and also wind farm controllability are being carried out. For example, GENERAL ELECTRIC in its 2.5 MW turbine specification [GE 2010] already offers the optimization of the wind power plant performance, which includes services such as, WindCONTROL* (Voltage and power regulation like a conventional plant) or WindFREE* (Provides reactive power even with no wind).
Transformers: Thanks to a variable transformer ratio, i.e., the relation between the primary and the secondary, the voltage can be controlled. As an example, a 400/220 kV transformer commonly has 21 tap positions (central tap ± 10), changing the voltage ± 1% with each tap modification.
In that sense, the tap position can be changed offline or online. This thesis focuses on the latter scheme, commonly known as OLTC transformer. In those cases a voltage set-point must be fixed. If the voltage measure differs from that set-point then the tap position is modified. That modification is not instantaneous; there is an intrinsic mechanical time delay which depends on two actions, the switching time and the motor drive. The former is around 50
P (kW)
Q (kVar)
200 kW 2000 kW
Cosφ =0.95 G87 - DFIG
Consumption Production
P (kW)
Cosφ =0.9
Full converter machine
Q (kVar) -650 kVar 650 kVar
Limited accuaracy Smax
ms for oil breaking transformers and 100 ms for vacuum transformers whereas the latter can be as high as 5 seconds. Nonetheless, some manufactures are trying to reduce this time as was seen at CIGRE bienal session [CIGRE. 2014] reaching mechanical time delays below 1 second. In addition, any OLTC transformer is defined by two additional settings, dead band and intentional time delay, which are used for coordination purposes. In this thesis as a relevant contribution those values are tuned for avoiding undesirable performance.
1.3.2.REGULATION FRAMEWORK OF THE WIND ENERGY VOLTAGE
CONTROL PROVISION3
Since the technical viability of providing reactive control has been proven, several countries have recently developed new grid codes, where some of them take into account the actual wind farms’ reactive power capabilities. The rise of wind farm requirements show a clear tendency: wind farm performance should be as similar as possible to the performance of a conventional power plant.
The grid codes of the European countries most representative from the wind share of total electricity perspective. Those countries are: Denmark, Spain, Portugal, Ireland and Germany [EWEA 2012] with 25.9%, 15.9%, 15.6%, 12% and 10.6%, penetration level respectively, being the mean value of the whole Europe 6.3%. Table 1-1 gathers the information indicating the TSO, the control mode specified, where the set-point is specified, the specifications, the time in which the set-point should be fulfilled, the voltage ranges and finally the corresponding reference. As can be seen in most countries, three different control modes are distinguished: reactive power, power factor, and voltage control. Currently the control mode that typically is used is the power factor. In fact, in countries like Spain or Portugal wind farms are operated at a unity power factor in order to reduce their impact on the TNet. Other countries such as Denmark or Ireland consider wider reactive ranges. Indeed, ENTSO-E wider range opens the doors of an enhanced utilization of the actual wind farm capabilities
System operators increasingly demand voltage control in the connection bus of the wind farm to the grid as is the case of Denmark, Ireland, Spain and also as has been proposed by ENTSO-E. In the event of considering this control model the system operator should define the droop (Voltage deviation / Reactive power increment). This concept, which is equivalent to the speed droop parameter R, is very well established in the primary control of active power (frequency control).
3 The regulation is subject to continuous changes. The regulation that is summarized in this document
Table 1-1. Reactive power technical requirements
As can be seen in Figure 1-4 a 4% droop means that a 4% frequency deviation is caused by a 100% change in the output power. In the event of voltage control, some countries use the equivalent of this definition, i.e., a droop of 4% means that a 4% of voltage deviation is caused by a 100% change in reactive power (Q lagging + Q leading). In the case of Spain, the droop is defined slightly differently: K= (∆V/Vrate) / (Q/Prate) where K goes from 0 to 25 corresponding to a droop of 4%. In all countries the system operator is the one in charge of fixing this value which is analyzed in this thesis from an optimal perspective.
0,1 lag - 0,1 lead Power
output range 11 kW -25 kW 30s 0.9 -1.06 0,228 lag - 0,228 lead Power
output range 25kW -25 MW 30s 0.9 -1.06 0,33 lag - 0,33 lead Power
output range >25 MW 30s
0.9 -1.06 (step1 0.9 -1.1) 0.995 lag - 0.995 lead Power
output range 11 kW -25 kW 30s 0.975 lag - 0.975 lead Power output range 25kW -25 MW 30s
0.95 lag - 0,95 lead Power output range >25 MW 30s Voltage control
(>25MW)
Required for wind plants higher than 25 MW 10s
0.9 -1.06 (step1 0.9 -1.1) Reactive power
control (*) 0,3 lag - 0,3 lead 0.95 - 1.05 0.979 lag - 0.979 lead
(mandatory) 0.995 lag - 0.995 lead
(incentivized)
Voltage control (*) 0-25 0.95 - 1.05 0.980 lag - 0.958 lead peak
hours (depends on the localization of the wind farms)
-1 valley hours -Reactive power
control 0,33 lag - 0,33 lead 20s Power factor control 0,95 lag - 0,95 lead 20s Voltage control 1%-10% 20s
0,228 lag - 0,48 lead 0,33 lag - 0,41 lead 0,41 lag - 0,33 lead 0,975 lag -0,887 lead 0,95 lag - 0,925 lead 0,925 lag - 0,95 lead Voltage control -Reactive power
control 0,5 lag - 0,65 lead -Power factor control 0,894 lag - 0,838 lead -Voltage control 2% -7%
-Europe ENTSO-E Grid connection point
[ENTSO-E 2012] [ENTSO-E 2013]
(**) The electricity undertaking choose the location of the voltage reference point. Normally at the high voltage side of the plant transformer. Inner envelope
0.9-1.075 Fixed outer envelope
0.875 -1.1 (*) Draft grid codes P.O. 12.2 & P.O. 7.5
380 kV: 0.92 – 1.16 220 kV: 0.88 – 1.15 110 kV: 0.87 – 1.15
[eon, 2006] Power factor control.
The TSO shall select one of the variants
[Ministério da economia,da inovação e do desenvolviento, 2010] Ireland EIRGRID, SONI Grid connection point Contingencies N-1 400 kV: 0,88 – 1,05 220 kV: 0,91 – 1,11 110 kV: 0,90 – 1,12
[REE, 1998], [REE, 2010], [REE, 2011]
[Spanish government, 2010]
Portugal REN Power factor control Grid connection point
Contingencies N-1 400 kV: 0,93 – 1,05 220 kV: 0,93 – 1,11 150 kV: 0,93 – 1,10 63 kV: 0,95 – 1,05 Power factor control
(Current scheme)
Contingencies N-1 400 kV: 0,95 – 1,09
(P.O. 1.1) [EirGrid, 2011], [EirGrid, 2014] Germany EON, EnBW, Vattenfall, RWE Reactive power control. The TSO shall select one of the
variants Grid connection
point 1 min
[Energinet, 2010] Power factor control
Spain REE Meter point 1 min
Reference Denmark Energinet Reactive power control Grid connection point (**)
Country TSO Control specified Set points
specified at Specifications
Set point changes completed
Voltage margins
400 kV: 0,8 – 1,15 220 kV: - – 1,12 150 kV: 0,9 – 1,13
63 kV: 0,9 – 1,12 0.69 kV: 0,9 – 1,10
Figure 1-4 Droop controls
For each control mode a different settling time may be defined, for example Denmark demands a faster settling time in the event of voltage control mode. The result is that a significant variability exists among countries (from 10 s to 1 minute). All set-points are commonly given at the grid connection point which typically corresponds to the high voltage side of the transformer. Nonetheless, in Denmark is chosen by the electricity supply undertaking company. In this case this point could be the grid connection bus, the common coupling bus bus (commonly selected if there is a tap changer) or one in between.
In Spain the set-point is specified at the meter point which can vary in location. In the event that all wind farms sharing the transformer belong to the same owner, the meter is at the high voltage side whereas if wind farms belong to different owners the meter point is located at the low voltage side. Moreover, in order to implement wind voltage control, it is necessary to coordinate the droop controls of wind farms and the current OLTC transformers control. This fact is outlined by Eirgrid and SONI (Ireland Transmission system operators). In order to define guidelines for adequate coordination different demonstrations are being carried out within the project “Delivering a Secure Sustainable Electricity System (DS3)”, [Eirgrid et al. 2012, Eirgrid et al. 2013].
In all cases analyzed, this service (reactive power/voltage control) is mandatory not contemplating any remuneration scheme. Nonetheless, some grid codes continue under review as the Spanish one. In this country, one of the main concerns of the Spanish wind association (AEE), is this issue.
1.4.
OBJECTIVES
This thesis aims to design a voltage control scheme which takes the maximum advantage of wind farm reactive power capabilities. TSOs seek that the whole HNet resembles a conventional plant which could be integrated in a hierarchical voltage control scheme. This strategy, as will be discussed in this thesis, benefits both TSO and wind farm owners. Nonetheless, the HNets infrastructure could differ significantly. Hence, the control limits, i.e., the maximum reactive power that can be injected or absorbed could change sifnificantly varying also the voltage margins. These limits will be used for determining whether the HNet is suitable for providing this service or not. In the case that it is not suitable, other strategies should be considered. In addition, how the strategy should be implemented must be considered. In this thesis a special emphasis is placed on
fREF
PG= PREF Power generation
S
y
st
em
f
re
q
u
en
cy
4%
100%
A group with a speed droop
of 4%
PG
VREF
QG= QREF = 0
reactive generation
V
o
lt
a
g
e
4%
100%
A group with a voltage droop of 4%
QG
Lagging reactive consumption Leading
simple control schemes which most of them do not require communication. In addition, in all cases additional investments are not necessary and hence, these strategies can directly be implemented.
In any case, the voltage control design does not end with its steady-state analysis requiring temporal evolution assessments (referred to in this thesis as dynamic analysis). Taking into account this temporal evolution the control performance can be optimized by paying special attention to OLTC transformers and the possible interactions among controllers that may appear. Consequently, two types of analysis are going to be carried out: steady-state and dynamic.
Once the global picture is clear, the concrete objectives are presented in the subsequent subsections.
1.4.1.STEADY-STATE ASSESMENT AND HNETS CLASSIFICATION
As has been mentioned, the first step is the evaluation of the HNet limits and potentials. On one hand, the limits are imposed by the maximum reactive power that the whole HNet is able to absorb from or inject to the transmission network. This maximum reactive power determines the maximum voltage variation that can be obtained. This variation highly depends on the strength of the bus, i.e., on its short circuit power. Hence both magnitudes: reactive power and voltage variation should be evaluated in detail. In that sense the well-known OPF will be used, being the reactive power at the HNet common coupling (- and +) maximization the objective function. This contrast with the common employment of the OPF for determining generator operational points in accordance with well-know objectives such as power losses or cost minimization. On the other hand, the potential that wind farms’ reactive power capabilities represent for both TSO and the HNet itself are also evaluated. In that sense, in addition to the reactive power and the voltage variation of power losses are also included. Those potentials will be compared with the ones obtained by common control schemes evaluating the improvements required for taking maximal advantage of wind farm capabilities. For carrying this analysis a bespoke tool which evaluates any HNet considering any objective function and imposing a certain control scheme will be developed.
Subsequently, the HNets should be characterized in accordance with their limits and potentials. Anticipating the results, three different types of HNets will be identified proposing for each one a different control strategy:
o Power losses minimization
o HNet minimum impact on TNet
o Pro-active voltage control
Moreover a different way of implementing each strategy will be discussed trying to minimize communication requirements and even avoid them.
1.4.2.OPTIMAL STATIC SETTINGS DESIGN
For each control strategy the optimal static settings (i.e., settings that impose the steady-state) should be determined. It is clear that in all cases those setting can be optimally evaluated each period. This approach will be followed in the event of a pro-active voltage control. In this case the optimal wind farms and OLTC transformers set-points are calculated each period by means of quadratic programming as will be detailed explained in Chapter 6. Nonetheless, this approach demands communication which not always is justified. In that sense others approaches more suitable for the other control strategies are discussed: data mining techniques for the power losses minimization and fit-and-forget static setting for the HNet minimum impact on TNet strategy. The former contributes with simple control rules of easy interpretability and which in some circumstances and for some devices can be used in a decentralized manner, totally avoiding communication. The latter contributes evaluating the optimal fixed settings for all set of scenarios. In that sense, it is guarantee that the settings are feasible under all circumstance and hence communication can also be avoided.
1.4.3.TIME VARIANT AND OFFLINE TUNING OF DYNAMIC SETTINGS
In addition to the static settings each control scheme is defined by some relevant dynamic settings (i.e., settings that affect the temporal evolution as time delays, time constants or dead band). Those settings will be tuned offline guarantying an optimal control performance and not requiring online updating. In that sense a control scheme simulator (developed in this thesis) will be used jointly with different metaheuristic algorithms minimizing relevant objectives such as:
o Tap changes minimization
o Voltage breaches minimization
o Transformers voltage set-point deviation minimization
o Oscillations minimization
This contrast with other techniques proposed in the literature and which have been explained in the subsequent chapter such as model predictive control or multi-agents. Techniques used for an online application which could demand two-way communication not available yet in the HNet.
1.5.
STRUCTURE OF THE DOCUMENT
This dissertation contains 7 chapters and 8 appendices. Chapter 2 presents the state-of-the-art of wind farms’ reactive power capabilities utilization in order to enhance the power system operation. Moreover the solutions presented in the literature from both steady-state and dynamic perspectives are gathered. Subsequently, Chapter 3 first outlines the technical benefits of voltage control provision from a technical perspective. Then, this same chapter expounds the steady-state analysis of HNets determining their potential and limits. Moreover, those HNets will be classified as one of three different types of HNet (A, B and C) and for each one a preferred strategy is selected.
Subsequently, the next three chapters focus on a specific type of HNet. Specifically, Chapter 4 focuses on type A where the strategy selected is power losses minimization. Then, Chapter 5 analyzes type B HNets where a minimum HNet impact on TNet strategy has been chosen. Next, Chapter 6 evaluates the pro-active voltage control strategy. Finally conclusions and future research are drawn in Chapter 7.
Concerning the appendices, the first one (Appendix A) presents the HNets characteristic and wind farms requirements. Subsequently, a brief summary of the reactive power voltage relationship is provided in Appendix B showing the importance of reactive power control provision. Then, Appendix C presents the sensitivity matrixes that are used in pro-active voltage control. Next, Appendix D analyzes actual wind farm data obtaining correlation factors and time evolution patterns. After that, Appendix E presents the simulator developed within this thesis in order to carry out the dynamic analyses. Moreover, the metaheuristics algorithms employed are summarized in Appendix F. Appendix G collects the detailed dynamic analysis of all control schemes evaluated within Chapter 5 and Appendix H compares these schemes with the pro-active voltage control strategy. Finally, Appendix I provides a simple analysis of an internal wind farm network, which throughout the thesis has been neglected.
CHAPTER 2
2.
STATE OF THE ART
This chapter presents research that has been carried out concerning wind farms voltage control, outlining the gaps which will be addressed by this thesis. At first the main research focus was evaluating the impact of wind generation on network voltages trying to minimize it at the planning stage. However, it was seen that the controllability of this generation was needed. Hence, great efforts have been done in this sense as will be seen. Recalling the title of this thesis, the focus is HNets (a typical scheme in countries with high wind penetration such as Ireland or Spain). Nonetheless, wind farms in many countries are commonly located in DNets and schemes such as the smart grids have been more investigated. Hence, those networks cannot be forgotten and a deep understanding of their operation should be acquired. In those networks special attention to the enhanced control schemes which consider reactive power capabilities of wind farms must be paid. Among all strategies the more suitable ones in accordance with the HNet potentials will be selected as is later explained in Chapter 3. One of these strategies is a pro-active voltage control. For that purpose a control of decentralized devices (OLTC transformers and wind farms) and its coordination should be carried out. Nonetheless, as has been outlined in the previous chapter, the control of decentralized elements is well understood in the TNet, where several hierarchical controls have been implemented. In addition, several alternatives are proposed focusing on DNet. This chapter classifies and studies the philosophy behind all controls for its later application to the HNets. It must be not forgotten that the final goal is integrating the HNet into an existing TNet hierarchical control.